DSI · decision-space integrity

Deployment · v0.2.1

Four supported ways to run it.

Every profile is local-first and stateless by default, and in the primary pattern makes no provider call. Pick one; the getting-started guide walks the first two. The package name is dsi-product; the import module is dsi. Supported Python: 3.10+.

Supported profiles

Core, API, dashboard, or container.

A · LIBRARY

Core library

In-process audit inside your own Python. No web server, no database, no provider call. Install the wheel, then dsi doctor and dsi validate-install. Bundled domain packs: career, finance, relationship; or supply your own expected points.

B · API

Local API sidecar

A small local HTTP API for shadow or offline audit — your system keeps the prompt/response flow. Endpoints: /health, /ready, /version, /v1/audit, /v1/control-loop, /v1/evidence. Binds to loopback by default; optional API-key protection.

C · DASHBOARD

Local dashboard

Paste a prompt + response, audit it, view and export evidence. Ships in the API extra (there is no separate dashboard extra). Stateless single-audit sandbox — not a multi-user or identity platform.

D · DOCKER

Docker sidecar

A reproducible local or customer-hosted container. Runs non-root, no external provider dependency. Publish to loopback on the host; inject an API key with an environment variable; stateless — no volumes required.

The default is loopback (127.0.0.1). Bind to a network interface only behind a reverse proxy you deliberately configure and control. See security & operations.

Not provided by this release

Stated plainly for evaluators — 0.2.1 does not include:

  • Kubernetes / Helm deployment
  • a managed cloud service or hosted SaaS
  • multi-tenant hosting
  • enterprise SSO / OIDC / RBAC
  • high availability
  • automatic model proxying
  • autonomous enforcement or response blocking

DSI runs in offline, shadow, and supervised-review modes. It does not block, route, or enforce.